jianying-headless: turn editing plans into editable JianYing drafts

The awkward last step of AI video workflows: an agent can “think” a cut through, but its output never lands in editing software — either you export a finished file and lose editability, or you click through the GUI shot by shot. jianying-headless fills exactly that gap: it turns a structured editing plan into an editable JianYing multi-track draft, and after editing, the native JianYing engine on your machine exports H.264/AAC MP4. Manual changes made inside JianYing don’t sync back to the plan, but the draft itself is fully editable in the app — that is the dividing line from “render the final video” tools. Created September 15, 2026, the repository counted 2,954 stars as of October 3, written in Python. The 视频号 account AI纪年 covered it on October 3.

Hypit original	Native JianYing export

From the README’s Hypit collaboration case: both strips sample the same real project at seconds 1, 8, 17, 28, 37 and 48. The project documents this as editable-handover, not visually lossless conversion — custom fonts, word-level color animation and shadows are not preserved.

Core features

• Editable draft generation: video segmentation, multi-track composition, speed, volume, picture-in-picture, subtitles and titles all land in the draft; local video, images, voiceover, music and sound effects can be imported, and static OTF/TTF fonts are saved with the draft. Linear keyframes cover position, scale, rotation, opacity and volume.
• Editing existing projects: inspects a source draft and modifies a separate copy, never overwriting the original — a safety design for engineering handoff.
• Native export: verified snapshots export to MP4 through the local JianYing engine. The README’s Hypit case: 23 tracks and 154 clips, passing a 1,507/1,507 frame check and full decode verification.
• Version and component checks: app version, build, official library hashes, signature and developer identity are all verified; unknown versions or mismatched components are rejected outright rather than run with loosened checks.
• Windows fallback: a community-contributed FFmpeg export path (PR #1) needs no JianYing and produces no JianYing draft — just an MP4, explicitly narrower than the native macOS flow.
• Agent Skill: the repo ships a standalone Agent Skill (a Python CLI entry) so editing can plug directly into agent workflows.

Typical use cases

• Engineering handoff in AI video pipelines: the upstream agent produces an editing plan; the downstream editor receives a multi-track JianYing project they can keep refining, not a locked file.
• Batch draft generation: talking-head clips, graphics and subtitles drafted in bulk, polished by hand.
• Agent-assisted editing: let the agent handle structural rough cuts and assembly, leaving keyframes and native effects to JianYing.

Quick start

Requires an Apple Silicon Mac (macOS 26.0+, verified on 26.5.1), JianYing Pro 11.5.0 (compatible 11.4.2), Python 3.9+, FFmpeg and Xcode Command Line Tools:

git clone https://github.com/mcncarl/jianying-headless.git
cd jianying-headless
python3 tools/build_native_codec.py
python3 skills/yichen-jianying-edit/scripts/headless_draft.py doctor

Run doctor first for environment and project checks; the official getting-started guide covers first install, importing media and common errors.

Summary

Get the licensing straight first: this is not conventional open source — the LICENSE is a custom personal learning and non-commercial use license. You may view, clone, study and modify it, but commercial use is off-limits and repackaging it as a skill collection or template bundle is prohibited; commercial use requires the author’s written permission. The 视频号 post called it a “public project” — the accurate phrasing is source-available, non-commercial. Second, version matching is a hard gate: the README states outright that 11.5.0 still requires matching install identity and toolchain, with no guarantee it runs on any given machine, and the official engine, account data and effect resources are not distributed with the source. Third, conversion is one-way — manual edits inside JianYing don’t flow back to the plan. Last push was September 27, 2026, so iteration is recent. If your workflow starts by collecting footage, pair it with yoinks, published here the same day — pull videos from across the web in your terminal, then feed the JianYing pipeline.